Is The Robot (AI) Apocalypse Near? – September NCTA General Meeting
by Sarah Schneider
On Saturday, September 9th, at the NCTA quarterly meeting, Carola F. Berger gave a presentation entitled “Is The Robot (AI) Apocalypse Near?”, addressing some very current and interesting topics in her discussion, and namely answering two questions: 1) Can machines think? and 2) Is this the end of humans?
First, we learned more about how Artificial Intelligence works, with Carola explaining that it’s essentially based on neural networks attempting to mimic the brain with complicated, interconnected nodes. The networks are trained with enormous amounts of data, which are assigned different weights and based on statistics, sequences, and patterns in order to produce output. Stanford has defined AI as, “a computer program making decisions based on patterns in data,” while Carola called it “statistics on steroids,” underlining that the most important thing to remember is that AI isn’t sentient, merely operating to regurgitate data.
Carola then pointed out the extensive and increasing investments that are being funneled into AI, rendering it a technology whose failure would be too costly. To put this into perspective, over 200 billion dollars were invested in AI in 2021 and 2022 alone, while just under 2 billion were invested in the machine translation market not related to generative AI. Another frightening statistic is the amount of CO2 emissions created by the computers running AI, which could eventually be on par with the amount of energy consumed by single countries.
To answer the first question – Can machines think? – scientists have elaborated various tests, of which Carola illustrated two. The Turing Test states that if a machine can have a conversation with a human without being detected as a machine, it has demonstrated human intelligence. In the Winograd Schema Challenge, AI is presented a brief discourse containing several aspects including ambiguous pronouns which a human could easily infer. While the AI engine managed to pass the first test, the second Winograd Schema test resulted in the AI’s hilarious failure.
One of the biggest underlying issues is how machines learn. The training costs for large language models grow exponentially, requiring trillions of parameters for an even remotely decent output.
And on to the second question: Are we going to be replaced by robots? Carola cleverly showed how AI requires humans for its instruction, demonstrating how if AI continues to be fed its own content, it indeed loses intelligence. One such example is MAD – Model Autophagy Disorder. In a vein much like Mad Cow Disease, when fed back its own generated images AI begins to produce new images of people with grids and stripes on their faces. This is just one small, yet noteworthy, example of how much AI needs humans to feed it original data.
Carola gave several examples highlighting the dangers of AI without a human filter. From an AI-generated book available on Amazon encouraging people to taste mushrooms to determine their toxicity, to algorithms that likely caused a flash crash of the British pound in an over-reaction to tweets, to a lawyer who used ChatGPT in courts and cited fake cases and for which the judge is considering sanctions. We also heard about Carola’s own personal experience asking ChatGPT about terminology in nuclear fusion and being provided several non-existent terms.
Even more frightening is the potential hidden bias in all sorts of things where one wouldn’t imagine AI even being involved; for example, when submitting a resume to a company, which is then incorrectly filtered out by AI. Another example is entirely incorrect insurance rates based on AI patterns that may be untrue.
As for AI in the translation and interpreting industry, we must absolutely educate clients on why we are still necessary. Carola is convinced that a certain segment of the translation market will disappear along with certain agencies that can be replaced by “robots,” i.e., those agencies that simply input material without intervention. In conjunction, other segments of the market will come to need us even more. AI won’t ever eliminate translators, as we provide added value.
The bottom line is that AI can be helpful, but humans are an imperative part of its past, present, and future. AI cycles improve and then worsen, and humans will always be needed to fix these matters.
Sarah Schneider is an ATA-certified Italian to English translator who lived near Bologna, Italy for over 15 years before moving back to the Bay Area in 2017 with her husband and two children. She specializes in the translation of patents and large sustainability reports, as well as tourism, marketing, and food & wine related documents.